How to save the compressed file in singular value decomposition

I am using the built-in function of svds for compressing image the problem is that i want to save the compressed file before reconstructing image the code is as the following

    function [im] = imcompr (infile,singvals,outfile)
    % Example of use:
    % out=imcompr('input_image.dcm',20,'output_image.dcm');
    % and [n,m]=size(input_image)
    if (exist(infile)==2)
   a = dicomread(infile);
   a = mat2gray(a);
      figure('Name','Input image');
   imshow(a);
end
if ndims(a)==2
    dvalue=double(a);
end
    [u,s,v] = svds(dvalue, singvals);
    if isa(a,'uint8')
        im = uint8(u * s * transpose(v));
    end
        if isa(a,'uint16')
        im = uint16(u * s * transpose(v));
    end
        if isa(a,'double')
        im = (u * s * transpose(v));
    end
   dicomwrite(im, outfile);
figure('Name','Output image');
imshow(im);
return;
end

[Walter Roberson]

By the way, you can simplify

if isa(a,'uint8')
        im = uint8(u * s * transpose(v));
    end
        if isa(a,'uint16')
        im = uint16(u * s * transpose(v));
    end
        if isa(a,'double')
        im = (u * s * transpose(v));
    end

to

    im = cast( u * s * transpose(v), class(a) );

[Walter Roberson]

Which of your variables contains the information you want to save?

[haithem abdelghany]

I want to save the variable which contain the compressed data before decompression but when i save s or the vector [u,s,v] it give me file which is more in value than the original image

[haithem abdelghany]

No here is the code http://www.advancedsourcecode.com/svdcompression.asp

Answer 1

[Walter Roberson]

Remember that dlmwrite() writes as text, and text is often about 3 times larger than binary. In order for the output text file to be smaller than the original binary file, the values would have to be really fortunate (e.g., 1 digit integers plus a comma could be smaller than the binary representation of the values), or else your output array would have to have fewer elements than the original to compensate for text taking up more room.

If you want to store s in a file in a way that does not expand like I describe above, you need to store it in binary, and you need to store as few bits for each element as you can get away with and still distinguish all the possible outputs.

For example, if you could happen to prove that the s array consists entirely of integers in the range -32768 to +32767 then you could store them as int16, and that might happen to be less than whatever was used to store them in the .dcm file. But I suspect you will find that you need at least single precision, 4 bytes per location, and that the result will be larger than the original matrix, which is probably int16 or uint16, 2 bytes per location. (Not every proposed compression technique works, and some compression techniques are only suitable for lossy compression situations.)

[haithem abdelghany]

But i tried a code for svd from site which is written in p-file which write the compressed file in .mat file and the compression ratio is about 12

[Walter Roberson]

It is a lossy compressor like I guessed.

http://fourier.eng.hmc.edu/e161/lectures/svdcompression.html

[haithem abdelghany]

No here is the code http://www.advancedsourcecode.com/svdcompression.asp

[Walter Roberson]

I am not going to pay them to get the code just to prove to you that svd as an image compression technique depends upon not storing all of the diagonal entries of the s matrices.

[haithem abdelghany]

I do not want you to pay for the code you miss understand me I downloaded the demo version it is written in p-function I mean that the code and the technique work well and I want to develop my code to work like it In other words i want to store the compressed data before decompression and i wish any one can help me

[Walter Roberson]

I wrote "It is a lossy compressor like I guessed." and you said "No", even though you have not seen the code yourself. I already posted a link to a paper that describes the theory of SVD compression and I could post additional links, but the only way that I could prove to you that they are in fact doing lossy compression would be for one of us to buy the code.

http://www.iosrjournals.org/iosr-jce/papers/Vol17-issue1/Version-2/B017120611.pdf

Notice the table near the end. For any lossless compression technique, MSE (Mean Squared Error) would be exactly 0.0, so any entry there with non-zero MSE is a lossy compression technique.

Anyhow, I've told you above what you should do:

"If you want to store s in a file in a way that does not expand like I describe above, you need to store it in binary, and you need to store as few bits for each element as you can get away with and still distinguish all the possible outputs."

[haithem abdelghany]

Thank you I understand